Abstract
Research in machine learning, mobile robotics, and autonomous driving is accelerated by the availability of high quality annotated data. Audi released the Audi Autonomous Driving Dataset (A2D2), consisting of simultaneously recorded images and 3D point clouds, together with 3D bounding boxes, semantic segmentation, instance segmentation, and data extracted from the automotive bus.
The sensor suite consists of six cameras and five LiDAR units, providing full 360° coverage. The recorded data is time synchronized and mutually registered. In addition, the dataset provides 392,556 sequential frames with registered LiDAR depth recorded across three cities in southern Germany, directly suitable for depth estimation research. Faces and vehicle number plates are blurred due to GDPR legislation. A2D2 was released by Audi under the CC BY-ND 4.0 license, permitting commercial use subject to the terms of the license.
Dataset Statistics
Data Access
The dataset is hosted on AWS S3 and is freely accessible without an AWS account.
Extracted dataset (recommended)
All files individually accessible in the canonical directory layout — browse sequences,
download single files or subsets, no huge archives to unpack. Integrity of every file can be
verified against the included CHECKSUMS.txt.
aws s3 ls --no-sign-request s3://audi-autonomous-driving-dataset/
AWS Region: eu-central-1 ·
ARN: arn:aws:s3:::audi-autonomous-driving-dataset
Original tar archives
The original distribution as monolithic tar archives:
aws s3 ls --no-sign-request s3://aev-autonomous-driving-dataset/
AWS Region: eu-central-1 ·
ARN: arn:aws:s3:::aev-autonomous-driving-dataset
Full documentation and download instructions: AWS Open Data Registry
License
A2D2 is released under the Creative Commons Attribution–NoDerivatives 4.0 (CC BY-ND 4.0) license, which permits commercial use subject to the terms of the license.
The vast majority of publicly available datasets are released under licenses permitting research-only use. Whilst we understand the reasons for this, we want to push progress in the field by publishing A2D2 under this less restrictive license. We hope this will help researchers, particularly those working within commercial enterprises.
A significant portion of cutting-edge research in autonomous driving and machine learning takes place inside private companies. We believe in permissive licensing as a means of fostering innovation broadly — and in contributing to startups and commercial entities by freely releasing data that is expensive to generate.